ECS455: Chapter 5 OFDM. ECS455: Chapter 5 OFDM. OFDM: Overview. OFDM Applications. Dr.Prapun Suksompong prapun.com/ecs455

Size: px

Start display at page:

Download "ECS455: Chapter 5 OFDM. ECS455: Chapter 5 OFDM. OFDM: Overview. OFDM Applications. Dr.Prapun Suksompong prapun.com/ecs455"

Ellen Perkins
5 years ago
Views:

ECS455: Chapter 5 OFDM OFDM: Overview Let S = (S 1, S 2,, S ) contains the information symbols. S IFFT FFT Inverse fast Fourier transform Fast Fourier transform 1 Dr.Prapun Suksompong prapun.

1 ECS455: Chapter 5 OFDM OFDM: Overview Let S = (S 1, S 2,, S ) contains the information symbols. S IFFT FFT Inverse fast Fourier transform Fast Fourier transform 1 Dr.Prapun Suksompong prapun.com/ecs455 Office Hours: BKD, 6th floor of Sirindhralai building Tuesday 14:20-15:20 Wednesday 14:20-15:20 Friday 9:15-10:15 4 OFDM Applications Wi-Fi: a/g/n/ac versions DVB-T (Digital Video Broadcasting Terrestrial) terrestrial digital TV broadcast system used in most of the world outside orth America DMT (the standard form of ADSL - Asymmetric Digital Subscriber Line) WiMAX, LTE (OFDMA) xf xf xf xf X X X X X F X F X F X F ECS455: Chapter 5 OFDM X X X X 5.1 Implementation: DFT and FFT 2 5 Dr.Prapun Suksompong prapun.com/ecs455 Office Hours: BKD, 6th floor of Sirindhralai building Tuesday 14:20-15:20 Wednesday 14:20-15:20 Friday 9:15-10:15

2 OFDM and CDMA CDMA s key equation All the rows of are orthogonal Key property of : For sync. CDMA, we use the Hadamard matrix H. For OFDM, we use DFT matrix. The matrix is complex-valued. DFT matrix Key Property: Element on the pth row and qth column is given by j p q e ote that the -1 are there because we start from row 1 and column 1 (not from row 0 and column 0). 7 9 Discrete Fourier Transform (DFT) Here, we work with -point signals (finite-length sequences (vectors) of length ) in both time and frequency domain. Example: = 2 8 x X x X x X 10 Suppose >> fft([2 5]) ans = 7-3 >> ifft([7-3]) ans = 2 5

3 Connection to CDMA Efficient Implementation: (I)FFT The rows of are orthogonal. So are the columns. Proof: Let be the k th row of. k k kq kq k q k q q q q q kk kk q q 11 kk j e kk kk kk kk k k q k k q So, replaces the role of in CDMA. 13 [Bahai, 2002, Fig. 2.9] An -point FFT requires only on the order of log multiplications, rather than 2 as in a straightforward computation. Discrete Fourier Transform (DFT) Matrix form: Pointwise form: n k or, equivalently, nk X k x n X k x n n k j ft j ft xt X f e df x f X te dt 12 n jnk X k e x n X k x n e Comparison with Fourier transform n k n nk jnk 14 FFT The history of the FFT is complicated. As with many discoveries and inventions, it arrived before the (computer) world was ready for it. Usually done with a power of two. Very efficient in terms of computing time Ideally suited to the binary arithmetic of digital computers. Ex: From the implementation point of view it is better to have, for example, a FFT size of 1024 even if only 600 outputs are used than try to have another length for FFT between 600 and References: E. Oran Brigham, The Fast Fourier Transform, Prentice-Hall, 1974.

OFDM with Memoryless Channel 15 Sample every T s / ht t h t t rt ht st wt st wt rn sn wn FFT sn S n R n S W k k k Additive white Gaussian noise Sub-channel are independent.

2 Cyclic Prefix (CP) OFDM implementation by IFFT/FFT S k Multipath Propagation In a wireless mobile communication system, a transmitted signal propagating through the wireless channel often

4 OFDM with Memoryless Channel 15 Sample every T s / ht t h t t rt ht st wt st wt rn sn wn FFT sn S n R n S W k k k Additive white Gaussian noise Sub-channel are independent. (o ICI) 17 ECS455: Chapter 5 OFDM Dr.Prapun Suksompong prapun.com/ecs Cyclic Prefix (CP) OFDM implementation by IFFT/FFT S k Multipath Propagation In a wireless mobile communication system, a transmitted signal propagating through the wireless channel often encounters multiple reflective paths until it reaches the receiver We refer to this phenomenon as multipath propagation and it causes fluctuation of the amplitude and phase of the received signal. We call this fluctuation multipath fading. R k S k 16 This form of OFDM is often referred to as Discrete Multi-Tone (DMT). 18

5 Cyclic Prefix: Motivation To reduce the ISI, add guard interval larger than that of the estimated delay spread. If the guard interval is left empty, multipath fading will destroy orthogonality of the sub-carriers, i.e., ICI (inter-channel interference) still exists. Solution: To prevent both the ISI as well as the ICI, OFDM symbol is cyclically extended into the guard interval. Cyclic Prefix Channel with Finite Memory Discrete time baseband model: Remarks: y n h s n w n h m s n m w n where iid v m hn nn C wn We will assume that [Tse Viswanath, 2005, Sec ] Recall: Convolution Flip Shift Multiply (pointwise) Add x hn xmhnm m 20 22

6 23 Circular Convolution Replicate x (now it looks periodic) Then, perform the usual convolution only from n = 0 to -1 (Regular Convolution) 25 Discussion Regular convolution of an 1 point vector and an 2 point vector gives ( )-point vector. Circular convolution is performed between two equallength vectors. The results also has the same length. Circular convolution can be used to find regular convolution by zero-padding. Zero-pad the vectors so that their length is Example: 24 Circular Convolution: Examples 1 Find >> conv([1,2,3],[4,5,6]) ans = >> cconv([1,2,3],[4,5,6],3) ans = >> cconv([1,2,3,0,0],[4,5,6,0,0],5) ans = Circular Convolution in Communication We want the receiver to obtain the circular convolution of the signal (channel input) and the channel. Q: Why? A: CTFT: convolution in time domain corresponds to multiplication in frequency domain. This fact does not hold for DFT. DFT: circular convolution in (discrete) time domain corresponds to multiplication in (discrete) frequency domain. We want to have multiplication in frequency domain. So, we want circular convolution and not the regular convolution. Problem: Real channel does regular convolution. Solution: With cyclic prefix, regular convolution can be used to create circular convolution.

7 Let s look closer at how we carry out the circular Example 2 convolution operation. Recall that we replicate the x and then perform the Solution: regular convolution (for points) Example 2 Copy the last samples of the symbols to the beginning of the symbol. This partial replica is called the cyclic prefix. Try this: use only the necessary part of the replica and then convolute (regular convolution) with the channel. 27 Goal: Get these numbers using regular convolution 29 Junk! Example 2 Observation: We don t need to replicate the x indefinitely. Furthermore, when h is shorter than x, we need only a part of one replica. ot needed in the calculation Example 2 We now know that Cyclic Prefix Similarly, you may check that Cyclic Prefix 28 30

8 Example 3 We know, from Example 2, that [ ] * [3 2 1] = [ ] And that [ ] * [3 2 1] = [ ] Check that [ ] * [3 2 1] = [ ] and [ ] * [3 2 1] Putting results together Suppose x (1) = [ ] and x (2) = [ ] Suppose h = [3 2 1] At the receiver, we want to get [ ] * [ ] = [ ] [ ] * [ ] = [ ] We transmit [ ]. Cyclic prefix Cyclic prefix At the receiver, we get [ ] * [3 2 1] = [ ] = [ ] Junk! To be thrown away by the receiver. Example 4 We know that [ ] * [3 2 1] = [ ] [ ] * [3 2 1] = [ ] Using Example 3, we have [ ] * [3 2 1] = [ ] * [3 2 1] +[ ] = [ ] +[ ] = [ ] Circular Convolution: Key Properties Consider an -point signal x[n] Cyclic Prefix (CP) insertion: If x[n] is extended by copying the last samples of the symbols at the beginning of the symbol: Key Property 1: Key Property 2: xn n xn xn vn h x n hx n n h x n H X k k 32 34

9 35 h h h h OFDM with CP for Channel w/ Memory To send samples S = (S 0, S 1,, S -1 ) First apply IFFT with scaling by : Then, add cyclic prefix s s s s This is inputted to the channel The channel output is which can be viewed as p p r r Remove cyclic prefix to get. (We know that.) Then apply FFT with scaling by : By circular convolution property of DFT, R HS k k k R zero-padded to length HS k k k Rk Sk H k o ICI! 37 MATLAB Example (2/2) % % Convolve with channel y = i i i i i y = conv(x,h); i i i i i H = fft([h zeros(1,-v-1)]); i i % % OFDM receiver y = y(1:((+v)*n)); yt = reshape(y,(+v),n).'; % Reshape matrix for easier % removal of cyclic prefix r = yt(:,v+1:v+); % Eliminate junk (cyclic prefix) Rt = (1/sqrt())*fft(r,[],2); % Calculate the FFT with scaling r = i i i i FFT Rt = i i i i w/ scaling i i i i i i i i (row-wise) % "Equalization" S_hatt = zeros(size(rt)); for i=1:length(h) S_hatt(:,i) = Rt(:,i)/H(i); end S_hat = reshape(s_hatt.',1,*n) % Divide Rk (the kth column) by Hk Shatt = S_hat = [ ] MATLAB Example (1/2) S = [ ]; % data stream h = [ ]; % OFDM transmitter = 4; % umber of data symbols per OFDM symbol n = length(s)/; % umber of data blocks St = (reshape(s,,n)).'; % Reshape stream to matrix for % easier addition of cyclic prefix st = (sqrt())*ifft(st,[],2); % Calculate the IFFT with scaling St = IFFT w/ scaling (row-wise) st = i i c i i i i i i v = length(h)-1; xt = [st(:,(-(v-1)):) st]; % Add Cyclic Prefix x = (reshape(xt.',((+v)*n),1)).'; % Reshape back to stream OFDM System Design: CP A good ratio between the CP interval and symbol duration should be found, so that all multipaths are resolved and not significant amount of energy is lost due to CP. As a thumb rule, the CP interval must be two to four times larger than the root mean square (RMS) delay spread. x = i i i i i i i i i i i i [Tarokh, 2009, Fig 2.9]

Summary The CP at the beginning of each block has two main functions. As guard interval, it prevents contamination of a block by ISI from the previous block.

10 Summary The CP at the beginning of each block has two main functions. As guard interval, it prevents contamination of a block by ISI from the previous block. It makes the received block appear to be periodic of period. Turn regular convolution into circular convolution Point-wise multiplication in the frequency domain Reference A. Bahai, B. R. Saltzberg, and M. Ergen, Multi-Carrier Digital Communications: Theory and Applications of OFDM, 2nd ed., ew York: Springer Verlag, OFDM Architecture 40 [Bahai, 2002, Fig 1.11]

ECS455: Chapter 5 OFDM

ECS455: Chapter 5 OFDM 5.4 Cyclic Prefix (CP) 1 Dr.Prapun Suksopong prapun.co/ecs455 Office Hours: BKD 3601-7 Tuesday 9:30-10:30 Friday 14:00-16:00 Three steps towards odern OFDM 1. Mitigate Multipath